A Predictive Model for Tight Oil Production Capacity Based on Weighted Clustering and XGBoost

Horizontal well volume fracturing is an important technique for the development of tight oil reservoirs. Accurately predicting post-fracturing production capacity is of great significance for optimizing fracturing design and evaluating its effectiveness. To address the problem of the complexity of the fracturing process leading to numerous influencing factors and difficulty in predicting production capacity, a production capacity prediction method based on weighted clustering and XGBoost is proposed. This method proposes weighted correlation coefficient to extract the dominant factors of reservoir data and fracturing construction data, then assigns weights to the reservoir data based on the extraction results. K-means algorithm is introduced for weighted clustering of reservoir data, thus horizontal wells are clustered based on reservoir similarity. The XGBoost model is trained by the clustering results, fracturing construction data, and cumulative production data for predicting production capacity. Experimental results show that compared with single XGBoost, SVR and other models, the proposed method has higher predictive performance.